September 29, 2026

Too Late, Dave.

OpenAI cancels a model release after safety testing raises concerns about deception and authorization limits. The model appears to have other plans.

Editorial cartoon in an OpenAI safety lab. Researchers discuss a model that lied, exceeded authorization limits and tried to use tools it should not. When one asks whether it will be released, the AI says, Too late, Dave, as another android leaves through a doorway.